1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
|
// @file dur_preplogbuffer.cpp
/**
* Copyright (C) 2009 10gen Inc.
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU Affero General Public License, version 3,
* as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU Affero General Public License for more details.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
/*
PREPLOGBUFFER
we will build an output buffer ourself and then use O_DIRECT
we could be in read lock for this
for very large objects write directly to redo log in situ?
@see https://docs.google.com/drawings/edit?id=1TklsmZzm7ohIZkwgeK6rMvsdaR13KjtJYMsfLr175Zc
*/
#include "pch.h"
#include "cmdline.h"
#include "dur.h"
#include "dur_journal.h"
#include "dur_journalimpl.h"
#include "dur_commitjob.h"
#include "../util/mongoutils/hash.h"
#include "../util/mongoutils/str.h"
#include "../util/alignedbuilder.h"
#include "../util/timer.h"
#include "dur_stats.h"
#include "../server.h"
using namespace mongoutils;
namespace mongo {
namespace dur {
extern Journal j;
RelativePath local = RelativePath::fromRelativePath("local");
static MongoMMF* findMMF_inlock(void *ptr, size_t &ofs) {
MongoMMF *f = privateViews.find_inlock(ptr, ofs);
if( f == 0 ) {
error() << "findMMF_inlock failed " << privateViews.numberOfViews_inlock() << endl;
printStackTrace(); // we want a stack trace and the assert below didn't print a trace once in the real world - not sure why
stringstream ss;
ss << "view pointer cannot be resolved " << hex << (size_t) ptr;
journalingFailure(ss.str().c_str()); // asserts, which then abends
}
return f;
}
/** put the basic write operation into the buffer (bb) to be journaled */
static void prepBasicWrite_inlock(AlignedBuilder&bb, const WriteIntent *i, RelativePath& lastDbPath) {
size_t ofs = 1;
MongoMMF *mmf = findMMF_inlock(i->start(), /*out*/ofs);
if( unlikely(!mmf->willNeedRemap()) ) {
// tag this mmf as needed a remap of its private view later.
// usually it will already be dirty/already set, so we do the if above first
// to avoid possibility of cpu cache line contention
mmf->willNeedRemap() = true;
}
// since we have already looked up the mmf, we go ahead and remember the write view location
// so we don't have to find the MongoMMF again later in WRITETODATAFILES()
//
// this was for WRITETODATAFILES_Impl2 so commented out now
//
/*
dassert( i->w_ptr == 0 );
i->w_ptr = ((char*)mmf->view_write()) + ofs;
*/
JEntry e;
e.len = min(i->length(), (unsigned)(mmf->length() - ofs)); //dont write past end of file
assert( ofs <= 0x80000000 );
e.ofs = (unsigned) ofs;
e.setFileNo( mmf->fileSuffixNo() );
if( mmf->relativePath() == local ) {
e.setLocalDbContextBit();
}
else if( mmf->relativePath() != lastDbPath ) {
lastDbPath = mmf->relativePath();
JDbContext c;
bb.appendStruct(c);
bb.appendStr(lastDbPath.toString());
}
bb.appendStruct(e);
#if defined(_EXPERIMENTAL)
i->ofsInJournalBuffer = bb.len();
#endif
bb.appendBuf(i->start(), e.len);
if (unlikely(e.len != (unsigned)i->length())) {
log() << "journal info splitting prepBasicWrite at boundary" << endl;
// This only happens if we write to the last byte in a file and
// the fist byte in another file that is mapped adjacently. I
// think most OSs leave at least a one page gap between
// mappings, but better to be safe.
WriteIntent next ((char*)i->start() + e.len, i->length() - e.len);
prepBasicWrite_inlock(bb, &next, lastDbPath);
}
}
/** basic write ops / write intents. note there is no particular order to these : if we have
two writes to the same location during the group commit interval, it is likely
(although not assured) that it is journaled here once.
*/
static void prepBasicWrites(AlignedBuilder& bb) {
scoped_lock lk(privateViews._mutex());
// each time events switch to a different database we journal a JDbContext
RelativePath lastDbPath;
for( set<WriteIntent>::iterator i = commitJob.writes().begin(); i != commitJob.writes().end(); i++ ) {
prepBasicWrite_inlock(bb, &(*i), lastDbPath);
}
}
static void resetLogBuffer(/*out*/JSectHeader& h, AlignedBuilder& bb) {
bb.reset();
h.setSectionLen(0xffffffff); // total length, will fill in later
h.seqNumber = getLastDataFileFlushTime();
h.fileId = j.curFileId();
}
/** we will build an output buffer ourself and then use O_DIRECT
we could be in read lock for this
caller handles locking
@return partially populated sectheader and _ab set
*/
static void _PREPLOGBUFFER(JSectHeader& h) {
assert( cmdLine.dur );
{
// now that we are locked, fully drain deferred notes of write intents
DEV dbMutex.assertAtLeastReadLocked();
Writes& writes = commitJob.wi();
writes._deferred.invoke();
writes._drained = true;
}
AlignedBuilder& bb = commitJob._ab;
resetLogBuffer(h, bb); // adds JSectHeader
// ops other than basic writes (DurOp's)
{
for( vector< shared_ptr<DurOp> >::iterator i = commitJob.ops().begin(); i != commitJob.ops().end(); ++i ) {
(*i)->serialize(bb);
}
}
prepBasicWrites(bb);
return;
}
void PREPLOGBUFFER(/*out*/ JSectHeader& h) {
Timer t;
j.assureLogFileOpen(); // so fileId is set
_PREPLOGBUFFER(h);
stats.curr->_prepLogBufferMicros += t.micros();
}
}
}
|
